Adaptive Routing for Damaged Networks

This paper proposes and examines adaptive routing algorithms for communication networks that are subject to damage. These algorithms route calls through the network when the network configuration is not fully known, and adaptively reorder the routing tables as they gather more information about the network configuration. (The path that a call follows in the network is determined by routing tables. When a call reaches a node, a routing table is consulted to find the next node to attempt.) We concentrate on learning mechanisms that reorder the routing tables in real-time. For example, the success-to-top mechanism moves the table entry that led to a successful connection of a call to the top of the routing table. Success-to-top leaves the relative order of the other entries in the routing table unchanged. Other possible schemes include failure-to-bottom (entries that lead to unsuccessful connection attempts are placed on the bottom of the list), and success-up-one (in which the successful entry is moved up by one in the routing table). Markov chain models are described for success-to-top and failure-to-bottom schemes. Analytical expressions for the steady-state probabilities are used to form measures for these two strategies. We compare these measures for a wide selection of blocking probabilities. Further, a simulation model is used to evaluate the merits of all three (and more) schemes. The simulation provides network measurements not available from the analytical model. The simulation also examines information sharing mechanisms in which a single call is used to change the routing tables at many nodes.